In his article, he highlights one of the severe disadvantages to merge:
The comment that he essentially made is that the downside of MERGE's handling of concurrency is that when you concurrently INSERT, so at the same time as you're executing the MERGE statement, there is another INSERT going on, then MERGE might not notice that. MERGE would go into its INSERT logic, and then it would get a unique violation.
This means that any individual row insert during the insert logic of the merge query can cause a unique violation and tank the entire merge query.
In my opinion, it would be better for merge to offer the functionality to simply ignore the rows that cause unique violation exceptions instead of tanking the whole query.